Voice-dialing system using adaptive model of calling behavior

Abstract

A method and apparatus for assisting voice-dialing using a model of an individual's calling behavior to improve recognition of an input name corresponding a desired telephone number. When the individual picks up a telephone, activity is initiated in a neural network model of the individual's calling behavior that predicts the likelihood that different numbers will be called, given such predictors as the day of the week and the time of day. The model is constructed by training the neural network with data from the user's history of making and receiving telephone calls. The auditory output from an automatic speech recognition system and the output from the user model are integrated together so as to select the number that is most likely to be the number desired by the speaker. The system can also provide automatic directory assistance, by speaking the number aloud rather than dialing it. In one version, the system is a personal directory for an individual maintained on that individual's personal computer. In another version, the system serves as a directory for a given physical or virtual site, with information about the institutional organization at the site in addition to individual calling histories used to track calling patterns and make predictions about the likelihood of calls within the site.

What is claimed is: 
     
       1. A method for assisting voice-dialing comprising the steps of: receiving voice input from a user representing a name corresponding to a desired telephone number;   selecting stored names that most closely match the voice input;   predicting a likelihood of the user calling different telephone numbers based on a model characterized by previous call information adjusted to reflect the user's calling behavior over a time period by applying weights associated with the telephone numbers such that recent calling behavior is favored over previous calling behavior; and   determining the desired telephone number according to the predicted likelihood of the user calling the telephone number corresponding to each selected name.   
     
     
       2. The method of claim 1, wherein the model of the user's calling behavior includes weighting determined from previous calls by the user to at least one of the telephone numbers, and wherein the determining step includes the substep of: applying the weighting to order telephone numbers corresponding to the selected names.   
     
     
       3. The method of claim 1, wherein the model of the user's calling behavior includes weightings determined from previous calls that the user received from at least one of the telephone numbers, and wherein the determining step includes the substep of: applying the weighting to order telephone numbers corresponding to the selected names.   
     
     
       4. The method of claim 1, wherein the determining step includes the substep of: specifying a set of the telephone numbers that most likely includes the desired telephone number.   
     
     
       5. The method of claim 1, wherein the determining step includes the substep of: ordering a set of the telephone numbers according to the predicted likelihood that each telephone number in the set is the desired telephone number.   
     
     
       6. The method of claim 4, wherein the specifying step includes the substeps of: prompting the user to select one of the telephone numbers from the set; and   initiating a telephone call to the selected telephone number.   
     
     
       7. The method of claim 1 further comprising the step of: dialing the desired telephone number.   
     
     
       8. The method of claim 1 further comprising the step of: outputting the desired telephone number in a manner perceptible to the user.   
     
     
       9. The method of claim 1, wherein the model of the user's calling behavior comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the predicting step includes the substep of: examining output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       10. The method of claim 1, wherein the model of the user's calling behavior comprises an adaptive model that is alterable based on the user's environment and actions with respect to telephone calls, and wherein the predicting step includes the substep of: examining output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       11. The method of claim 1, wherein the model of the user's calling behavior comprises a neural network and wherein the predicting step includes the substep of: examining output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       12. The method of claim 1, wherein a speech recognition system receives the voice input, and wherein the determining step includes the substep of: integrating output of the calling behavior model with a second model of measures of confidence of the speech recognition system corresponding to the telephone numbers.   
     
     
       13. The method of claim 12, wherein the second model includes integration factors determined from previous calls by the user to at least one of the telephone numbers, and wherein the integrating step includes the substep of: applying the integration factors to select the desired telephone number.   
     
     
       14. The method of claim 13, wherein the applying step includes the substeps of: generating a set of the telephone numbers associated with the selected names; and ordering the set of the telephone numbers in accordance with the integration factors with telephone numbers determined more likely to be the desired telephone number ahead of telephone numbers determined less likely to be the desired telephone number.   
     
     
       15. The method of claim 12, wherein the second model comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the integrating step includes the substep of: examining output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       16. The method of claim 12, wherein the second model comprises an adaptive model that is alterable based on the user's environment and actions with respect to telephone calls, and wherein the integrating step includes the substep of: examining output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       17. The method of claim 12, wherein the second model of integration data is a neural network, and wherein the integrating step includes the substep of: examining output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       18. The method of claim 12, wherein the second model is a fixed procedure, and wherein the integrating step includes the substep of: combining information related to the telephone numbers corresponding to the selected names with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names.   
     
     
       19. The method of claim 1, further comprising the step of: training the model of the user's calling behavior with data corresponding to previous calls.   
     
     
       20. The method of claim 12, further comprising the step of: training the second model with confidence measures for the speech recognition system corresponding to previous calls.   
     
     
       21. The method of claim 4, wherein the generating step includes the substeps of: selecting a name associated with one of the telephone numbers in the set;   presenting the selected name to the user; and   waiting for a response from the user indicating whether the selected name corresponds to the desired telephone number.   
     
     
       22. The method of claim 21, wherein the waiting step includes the substep of: determining whether a predetermined period of time has passed since the user was presented with the selected name; and   interpreting a lack of response within the predetermined period as meaning that the selected name corresponds to the desired telephone number.   
     
     
       23. The method of claim 1, further comprising the step of: building a training set including information related to at least one previous call.   
     
     
       24. The method of claim 23, further comprising the step of: at a predetermined time, modifying the model of the user's calling behavior in accordance with the training set.   
     
     
       25. The method of claim 24, wherein the model of the user's calling behavior includes weightings determined from at least one previous call by the user to at least one of the telephone numbers, and wherein the modifying step includes the substep of: altering the weightings of the user's calling behavior model to reflect the information related to the previous call.   
     
     
       26. The method of claim 24, wherein the training set includes information related to a plurality of previous calls, and wherein the modifying step includes: iteratively altering the weighting of the user's calling behavior model to reflect the information related to the plurality of previous calls.   
     
     
       27. The method of claim 12, further comprising the step of: building a training set including information on at least one previous call and a confidence measure of the speech recognition system in selecting a stored name corresponding to the previous call.   
     
     
       28. The method of claim 27, further comprising the step of: at a predetermined time, modifying the second model in accordance with the training set.   
     
     
       29. The method of claim 28, wherein the modifying step includes that substep of: altering the second model in accordance with the confidence measure of the previous call in the training set.   
     
     
       30. Apparatus comprising: a receiver configured to receive voice input from a user representing a name corresponding to a desired telephone number;   a selector configured to select stored names that most closely match the voice input;   a predictor configured to predict a likelihood of the user calling different telephone numbers based on a model characterized by previous call information adjusted to reflect the user's calling behavior over a time period by applying weights associated with the telephone numbers such that recent calling behavior is favored over previous calling behavior; and   a determiner configured to determine the desired telephone number according to the predicted likelihood of the user calling the telephone number corresponding to each selected name.   
     
     
       31. The apparatus of claim 30, wherein the model of the user's calling behavior includes weighting determined from previous calls by the user to at least one of the telephone numbers, and wherein the determiner includes: applying means configured to apply the weighting to order telephone numbers corresponding to the selected names.   
     
     
       32. The apparatus of claim 30, wherein the model of the user's calling behavior includes weighting determined from previous calls that the user received from at least one of the telephone numbers, and wherein the determiner includes: applying means configured to apply the weighting to order telephone numbers corresponding to the selected names.   
     
     
       33. The apparatus of claim 30, wherein the determiner includes: specifying means configured to specify a set of the telephone numbers that most likely includes the desired telephone number. 
     
     
       34. The apparatus of claim 30, wherein the determiner includes: ordering means configured to order a set of the telephone numbers according to the predicted likelihood that each telephone number in the set is the desired telephone number.   
     
     
       35. The apparatus of claim 33, wherein the specifying means includes: prompting means configured to prompt the user to select one of the telephone numbers from the set; and   initiator configured to initiate a telephone call to the selected telephone number.   
     
     
       36. The apparatus of claim 30 further comprising: a dialer configured to dial the desired telephone number.   
     
     
       37. The apparatus of claim 30 further comprising: outputting means configured to output the desired telephone number in a manner perceptible to the user.   
     
     
       38. The apparatus of claim 30, wherein the model of the user's calling behavior comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the predictor includes: examining means configured to examine output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       39. The apparatus of claim 30, wherein the model of the user's calling behavior comprises an adaptive model that is alterable based on the user's environment and actions with respect to telephone calls, and wherein the predictor includes: examining means configured to examine output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       40. The apparatus of claim 30, wherein the model of the user's calling behavior comprises a neural network and wherein the predictor includes: examining means configured to output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       41. The apparatus of claim 30, wherein the receiver includes a speech recognition system and wherein the determiner includes: an integrator configured to integrate output of the calling behavior model with a second model of measures of confidence of the speech recognition system corresponding to the telephone numbers.   
     
     
       42. The apparatus of claim 41, wherein the second model includes integration factors determined from previous calls by the user to at least one of the telephone numbers and wherein the integrator includes: applying means configured to apply the integration factors to select the desired telephone number.   
     
     
       43. The apparatus of claim 42, wherein the applying means includes: generator configured to generate a set of the telephone numbers associated with the selected names; and   ordering means configured to order the set of the telephone numbers in accordance with the integration factors with telephone numbers determined more likely to be the desired telephone number ahead of telephone numbers determined less likely to be the desired telephone number.   
     
     
       44. The apparatus of claim 41, wherein the second model comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the integrator includes: examining means configured to examine output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       45. The apparatus of claim 41, wherein the second model comprises an captive model that is alterable based on the user's environment and actions with aspect to telephone calls, and wherein the integrator includes: examining means configured to examine output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       46. The apparatus of claim 41, wherein the second model of integration data is a neural network, and wherein the integrator includes: examining means configured to examine output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       47. The apparatus of claim 41, wherein the second model is a fixed procedure, and wherein the integrator includes: combining means configured to combine information related to the telephone numbers corresponding to the selected names with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names.   
     
     
       48. The apparatus of claim 30, further comprising: a trainer configured to train the model of the user's calling behavior with data corresponding to previous calls.   
     
     
       49. The apparatus of claim 41, further comprising: a trainer configured to train the second model with confidence measures for the speech recognition system corresponding to previous calls.   
     
     
       50. The apparatus of claim 33, wherein the generator includes: selecting means configured to select a name associated with one of the phone numbers in the set;   presenting means configured to present the selected name to the user; and   a monitor configured to wait for a response from the user indicating whether the elected name corresponds to the desired telephone number.   
     
     
       51. The apparatus of claim 50, wherein the monitor includes: determining means configured to determine whether a predetermined period of time has passed since the user was presented with the selected name; and   an interpreter configured to interpret a lack of response within the predetermined period as meaning that the selected name corresponds to the desired telephone number.   
     
     
       52. The apparatus of claim 30, further comprising: a training set builder configured to building a training set including information related to at least one previous call.   
     
     
       53. The apparatus of claim 52, further comprising: a modifier configured to, at a predetermined time, modify the model of the user's calling behavior in accordance with the training set.   
     
     
       54. The apparatus of claim 53, wherein the model of the user's calling behavior includes weightings determined from at least one previous call by the user to at least one of the telephone numbers, and wherein the modifier includes: altering means configured to alter the weightings of the user's calling behavior model to reflect the information related to the previous call.   
     
     
       55. The apparatus of claim 53, wherein the training set includes information related to a plurality of previous calls, and wherein the modifier includes: altering means configured to iteratively alter the weighting of the user's calling behavior model to reflect the information related to the plurality of previous calls.   
     
     
       56. The apparatus of claim 41, further comprising: a training set builder configured to build a training set including information on at least one previous call and a confidence measure of the speech recognition system in selecting a stored name corresponding to the previous call.   
     
     
       57. The apparatus of claim 56, further comprising: a modifier configured to modify at a predetermined time the second model in accordance with the training set.   
     
     
       58. The apparatus of claim 57, wherein the modifier includes: altering means configured alter the second model in accordance with the confidence measure of the previous call in the training set.   
     
     
       59. A method for initiating telephone calls by voice, comprising the steps of: activating, in response to a determination that a user intends to initiate a new call, a calling behavior model to predict a likelihood of the user intending to call each one of a predetermined set of telephone numbers, the calling behavior model being characterized by previous call information adjusted to reflect the user's calling behavior over a time period by applying weights associated with the telephone numbers such that recent calling behavior is favored over previous calling behavior;   receiving a voice input including a sequence of sounds that represent a name spoken by the user and corresponding to a telephone number that the user desires to call;   selecting names from a personal directory including voice data representing names associated with the predetermined set of telephone numbers, based on a comparison of the sequence of sounds from the voice input and voice data of the personal directory; and   integrating the selection of telephone numbers corresponding to the selected names from the personal directory with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names from the personal directory to identify the telephone number most likely to be the desired telephone number.   
     
     
       60. The method of claim 59, wherein the activating step includes the substep of: stimulating a category-based calling behavior model to predict a likelihood of the user intending to call each one of a predetermined set of institutional directory telephone numbers, and wherein the integrating step includes the substep of:   joining the predictions of the likelihood of the user intending to call each one of the institutional directory telephone numbers with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names from the personal directory.     
     
     
       61. The method of claim 60, wherein the category-based calling behavior model comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the joining step includes the substep of: examining output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       62. The method of claim 60, wherein the category-based calling behavior model comprises an adaptive model alterable in response to the user's environment and actions with respect to telephone calls, and wherein the joining step includes the substep of: examining output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       63. The method of claim 60, wherein the category-based calling behavior model comprises a neural network, and wherein the joining step includes the substep of: examining output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       64. The method of claim 60, further comprising the step of: training the category-based calling behavior model with historical calling data based on previous calls among institutional directory telephone numbers.   
     
     
       65. An apparatus comprising: an activator configured to activate, in response to a determination that a user intends to initiate a new telephone call, a calling behavior model to predict a likelihood of the user intending to call each one of a predetermined set of telephone numbers, the calling behavior model being characterized by previous call information adjusted to reflect the user's calling behavior over a time period by applying weights associated with the telephone numbers such that recent calling behavior is favored over previous calling behavior;   a receiver configured to receive a voice input including a sequence of sounds that represent a name spoken by the user and corresponding to a telephone number the user desires to call;   a selector configured to select names from a personal directory including voice data representing names associated with the predetermined set of telephone numbers, based on a comparison of the sequence of sounds from the voice input and voice data of the personal directory; and   an integrator configured to integrate the selection of telephone numbers corresponding to the selected names from the personal directory with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names from the personal directory to identify the telephone number most likely to be the desired telephone number.   
     
     
       66. The apparatus of claim 65, wherein the activator includes: a stimulator configured to stimulate a category-based calling behavior model to predict a likelihood of the user intending to call each one of a predetermined set of institutional directory telephone numbers, and wherein the integrator includes:   a joiner configured to join the predictions of the likelihood of the user intending to call each one of the institutional directory telephone numbers with the predictions of the likelihood of the user calling each of the telephone numbers associated with the selected names from the personal directory.   
     
     
       67. The method of claim 66, wherein the category-based calling behavior model comprises an abstract representation based on the user's environment and actions with respect to telephone calls, and wherein the joiner includes: examining means configured to examine output of the abstract representation for indications that the user intends to call each of the telephone numbers.   
     
     
       68. The apparatus of claim 66, wherein the category-based calling behavior model comprises an adaptive model alterable in response to the user's environment and actions with respect to telephone calls, and wherein the joiner includes: examining means configured to examine output of the adaptive model for indications that the user intends to call each of the telephone numbers.   
     
     
       69. The apparatus of claim 66, wherein the category-based calling behavior model comprises a neural network, and wherein the joiner includes: examining means configured to examine output of the neural network for indications that the user intends to call each of the telephone numbers.   
     
     
       70. The apparatus of claim 66, further comprising: a trainer configured to train the category-based calling behavior model with historical calling data based on previous calls among institutional directory telephone numbers.   
     
     
       71. A method for assisting voice-dialing comprising the steps of: receiving voice input from a user representing a name corresponding to a desired telephone number;   selecting stored names that most closely match the voice input;   predicting a likelihood of the user calling different telephone numbers based on a model of the user's calling behavior by applying weights associated with the telephone numbers such that recent calling behavior is favored over previous calling behavior; and   determining the desired telephone number according to the predicted likelihood of the user calling the telephone number corresponding to each selected name.